Genetic Programming Theory and Practice II (Genetic Programming, Volume 8)
Format: PDF / Kindle (mobi) / ePub
This volume explores the emerging interaction between theory and practice in the cutting-edge, machine learning method of Genetic Programming (GP). The contributions developed from a second workshop at the University of Michigan's Center for the Study of Complex Systems where leading international genetic programming theorists from major universities and active practitioners from leading industries and businesses met to examine how GP theory informs practice and how GP practice impacts GP theory. Chapters include such topics as financial trading rules, industrial statistical model building, population sizing, the roles of structure in problem solving by computer, stock picking, automated design of industrial-strength analog circuits, topological synthesis of robust systems, algorithmic chemistry, supply chain reordering policies, post docking filtering, an evolved antenna for a NASA mission and incident detection on highways.
both the initial BB supply model and the decision-making model in the populationsizing relation. They also eliminated the requirement that only a successful decision-making in the first generation results in the convergence to the optimum. Specifically, Harik et al., modeled the decision-malung in subsequent generations using the well known gambler's ruin model (Feller, 1970). The gambler's ruin population-sizing model was subsequently extended for noisy environments (Miller, 1997), and for
Advances in Genetic Programming, 75-97. Cambridge: The MIT Press. Angeline, P. J. (1997). Parse Trees. In T. Back, D. B. Fogel and Z. Michalewicz (Eds.), Handbook of Evolutionary Computation, C1.6:l-C1.6:3. Bristol: Institute of Physics Publishing. Banzhaf, W. and W. B. Langdon (2002). Some Considerations on the Reason for Bloat. Genetic Programming and Evolvable Machines, 3(1), 81 - 91. Banzhaf, W., P. Nordin, et al. (1998). Genetic Programming: An Introduction: On the Automatic Evolution of
(probability) of mutations and crossovers, contrary to most literature that we've read. We did find that there were some reasonable levels that allowed convergence with fewer generations; it turns out they were awfully close to the genetic program library defaults. The population sizes and number of demes certainly had impact on the diversity of the initial formulae that were built - generally the higher the better if you have time. Migration wait is basically a parameter that controls how long
Fitness,,, a Tree-Root I R C R I R Figure 9-3. An example of a GP tree, composed of topology operators applied to an embryo, generating a bond graph after depth-first execution (numeric branches are omitted) Topological Synthesis of Robust Dynamic Systems 149 Sustainable Genetic Programming Based on the Hierarchical Fair Competition Model Standard genetic programming has a strong tendency toward premature convergence of the GP tree structures, as illustrated by the visualization of GP
10-4. Design Evolution for Santa Fe Trail 11 IfFood 11 Final Design I ~ r o g 2I I ~ F O O ~ 70% ~rog3 1 15% 1 14% Tables 10-4-10-8 show a more detailed breakdown of the results of the third experiment. Each row in these tables shows the distribution of final designs given the initial design of the population. For example, in Table 10-4, the first row of data shows that with the Santa Fe Trail problem in the initial populations where I f Food is the design of the best individual, 70% of